Patients’ and Providers’ Attitudes Toward Artificial Intelligence and Electronic Health Record Use in Deep Phenotyping and Rare-Disease Screening: An Empty Systematic Review

Abstract

Background: The integration of Artificial Intelligence (AI) and algorithms into healthcare is transformative, particularly in diagnosing rare diseases (RDs), enhancing the accuracy and speed of condition identification. Objectives: This systematic literature review investigates perceptions and attitudes toward the use of AI in Electronic Health Records (EHRs) for screening patients at risk of RD, aiming to understand patients’ and healthcare providers’ expectations and concerns. Methods: Following PRISMA guidelines, a systematic search was performed in December 2023. A search strategy developed by the research team in collaboration with an expert librarian, using the PICO framework, was applied. Searches were conducted in PubMed, Scopus, and Web of Science. The search strategy covered four main concepts: diagnostic techniques, medical records, AI, and attitudes toward these technologies. Results: The initial search retrieved 3348 articles after duplicate removal. However, no studies met the inclusion criteria. As a result, no eligible studies were identified, preventing risk-of-bias assessment or data synthesis. Discussion: The absence of relevant studies highlights the need for further research focusing on patient and healthcare provider attitudes toward AI-integrated EHRs, especially in RD and their early detection. Conclusions: The lack of studies on stakeholder attitudes toward AI in EHRs for RD screening represents an important research gap. Addressing this gap will improve the understanding and development of AI applications in healthcare, ensuring they meet user needs and ethical standards.

1. Introduction

Perspectives from all stakeholders, including patients, healthcare providers, caregivers, and governments, have become increasingly important in medical research since the advent of the PREFER recommendations [1]. For example, in 2021, the World Health Organization (WHO) emphasized the need to integrate patient attitudes into medical development; rare diseases (RDs) are no exception [2,3]. It is important that research meaningfully involves users and providers to support the implementation of findings in clinical settings with the highest probability of success, especially in RD [4]. In recent years, the integration of Artificial Intelligence (AI) into the healthcare sector has represented a transformative shift, particularly in the screening and diagnosis of RDs. These technological advancements offer unprecedented opportunities to enhance the accuracy, efficiency, and speed of diagnosing complex conditions that are often difficult to identify [5]. Rare diseases, characterized by their low prevalence among the population (fewer than 1 in 2000 people) [6], present significant challenges for traditional healthcare methodologies due to their wide variety and overlap with symptoms of common diseases, as well as the heterogeneity of symptom presentation.

AI systems, through machine learning and deep learning algorithms (e.g., natural language processing), can analyze vast datasets far beyond the capacity of human practitioners, identifying patterns and correlations that may elude even the most experienced professionals (see definitions in [7]). This capability is crucial for RDs, where the limited number of cases can hinder the accumulation of comprehensive knowledge through experience alone. By leveraging data from electronic health records (EHRs), genetic information, and even wearable health devices, the use of algorithms in screening processes can streamline the identification of RDs, reducing the time and resources typically required [8]. Reducing time to diagnosis is particularly important given the diagnostic journey many patients with RDs face; many undergo years of misdiagnoses and ineffective treatments. By enhancing diagnostic accuracy and efficiency, AI/ML not only improves patient outcomes but also contributes to a better understanding of RDs, which may facilitate research into new treatments and interventions [9].

International research efforts are focusing on improving screening for RDs in newborns and advancing sequencing technologies. Large initiatives such as the Baby Seq project [10] and the Screen4Care (S4C) project aim to expedite RD diagnosis using genetic testing and digital technologies, including AI. The S4C project (the funding source for this research), for example, is funded by the European Union via an Innovative Medicines Initiative (IMI) project, a global initiative accelerating access to innovative medicines. S4C employs two key strategies: genetic newborn screening (gNBS), targeting selected RDs with treatable or actionable conditions, and innovative digital solutions designed to foster connectivity within RD communities and promote stakeholder involvement for expedited diagnosis [11].

However, the implementation of these technologies is not without challenges. Issues such as data privacy, the need for large and diverse datasets aligned with the FAIR (Findable, Accessible, Interoperable, Reusable) principles to train AI models [12,13], and the integration of AI tools into clinical practice must be addressed [14]. From the perspective of different stakeholders, AI and ML innovations are increasingly viewed as more reasonable, feasible [15], and acceptable [16].

In recent years, preference studies have been developed in the medical and health sectors to assess the impact of users’ and healthcare providers’ experiences. In RDs, the plausibility of AI and ML as aids for detection is increasingly a reality [8,17]. However, ethical issues are more pressing in RDs, as little information can create challenges because of data fragmentation, data quality, and data privacy (e.g., the potential identification of patients due to the “rare” aspect of their diagnosis) [18]. In an area of such high stakes and innovative growth, patient engagement and the inclusion of stakeholder attitudes are warranted and necessary.

To better contextualize the gaps identified in the review, it is important to note that several adjacent research domains are already well developed (particularly over the last 5 years) but remain largely disconnected. Systematic reviews have documented substantial progress in AI-based rare-disease detection and phenotyping using EHR data, particularly through machine learning and natural language processing approaches [19,20]. In parallel, a mature body of literature exists on EHR-driven phenotyping methods, focusing primarily on technical model development and evaluation [21,22]. Separately, stakeholder attitudes toward health data sharing have been extensively synthesized, highlighting concerns about privacy, trust, and secondary data use [23,24]. Finally, a growing number of systematic reviews address acceptance and trust in clinical AI tools, emphasizing the importance of user perspectives for successful implementation [25,26,27].

This literature review aims to investigate the perceptions and attitudes of stakeholders (including healthcare professionals, patients, and the general public) toward the use of AI/ML in electronic health records (EHRs) for screening RDs.

2. Methods

This review was performed in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. The review protocol was registered in the Open Science Framework (OSF) database (Registry number: 10.17605/OSF.IO/SDXP9).

2.1. Search Strategy

Together with a university librarian from Uppsala University, three researchers with diverse backgrounds (public health, clinician medicine, and psychology) developed the search strategy in December 2023. Databases were selected to cover both social sciences and medical literature (PubMed, Scopus, and Web of Science). Search terms were developed to encompass the four main concepts: diagnosis and diagnostic techniques, electronic health records, artificial intelligence, and attitudes toward these technologies. The initial search strategy included rare-disease-specific terms with no positive results on the selection criteria. Out of the 260 articles retrieved, none were eligible after title and abstract screening. We hypothesized that the rare-disease terms were too restrictive to capture the subtility of certain rare-disease names and genetic conditions (see Table A1). Therefore, we extended the search strategy while ensuring that all articles retrieved by the initial search were also included in the revised strategy, demonstrating that the expanded search was coherent and did not deviate from the RDs’ associated terms. The final search combined MeSH terms and keywords in titles and abstracts to identify relevant literature.

For the diagnosis and diagnostic techniques domain, the search included terms such as “Diagnosis”[MeSH] OR “diagnosis code*”, “phenotyp*”. The medical records domain included terms such as “medical records”[MeSH Terms] OR “Registries”[MeSH Terms]. The role of Artificial Intelligence was explored using “Artificial Intelligence”[MeSH], supplemented with keywords such as “algorithm*”, “artificial intelligence” in titles and abstracts. Attitudes toward these technologies were investigated using “Attitude”[Mesh] OR “Patient Preference”[MeSH] (see the full search strategy in Table A2). The search strategy employed logical operators to combine these concepts, ensuring that articles retrieved were relevant across all domains.

Professional librarians independently reviewed the search strategy to ensure consistency of key words and MeSH terms across the different databases. This approach was designed to capture a comprehensive set of articles at the intersection of AI, healthcare data management, and the human factors influencing the adoption and effective implementation of these technologies in healthcare settings (see search strategy in Table A2). The methods used in this review followed current guidelines and recommendations [28] (see Figure 1 for the PRISMA Flowchart). In addition, the reporting structure was informed by previously published systematic reviews with null findings [29].

2.2. Eligibility Criteria

To be included in the study, studies must have met all the following criteria:Full text was available.Written in English.Reported empirical research (quantitative or qualitative or mixed methods).Included stakeholders of interest (patients, the general public, or healthcare professionals).Addressed all four of the following concepts: ○(1) Deep learning or algorithms;○(2) Detection of diagnosis, diseases, or at-risk alerts for RD;○(3) Use of Electronic Health Records;○(4) Consideration of attitudes or preferences of users or healthcare providers.

All RDs were considered according to the European definition adopted by Orphanet, which states “In Europe, a disease is considered to be rare when it affects 1 (or less) person per 2000” [6]. The review included both early-onset RDs and ultra-RDs. This broad approach was intended to capture a wide range of research for inclusion, including all RDs that may be suitable for screening using digital health technologies. Studies without abstracts were included during the initial screening stage.

The exclusion criteria were:Conference and study protocols;Theoretical papers;Editorials;Research on animals;Studies focused on medical procedures;Technical development of AI/ML;Performance evaluations of diagnostic tools;Studies focused on treatment decisions or medication changes.

This approach ensured a thorough review of the literature relevant to technology’s role in healthcare, emphasizing the significant impact of Electronic Health Records and AI on healthcare delivery and perceptions (see Figure 2, Inclusion–Exclusion table). We therefore restricted the inclusion timeline to articles published from 2010 to 2024 for double blind review. However, articles published before 2010 were also screened by a single reviewer to ensure that no relevant paper was misclassified.

2.3. Selection Process

After the initial search, 4415 records were identified. Following the removal of duplicates, 3358 articles remained for screening. Using the rayyan.ai software (free access version without AI companion), all articles were independently screened in double-blind manner based on their titles and abstracts by four reviewers (SM, ÅG, BA, JC). Title and abstract screening adhered to the predefined inclusion and exclusion criteria, aimed at identifying studies assessing participants’ attitudes or preferences regarding the use of Electronic Health Records and Artificial Intelligence. Final decisions on inclusion and exclusion were reached through consensus among at least three reviewers, with final confirmation by the principal investigator to ensure that all selection was harmonized across reviewers’ competences and experiences. Because one reviewer was a medical doctor, articles with technical medical vocabulary were referred to that reviewer to support the final decision.

After title and abstract screening, 11 articles remained and underwent full-text review. Full-text screening was also conducted independently by two reviewers using the same consensus process. Ultimately, none of the studies were included in the results because they did not meet the inclusion criteria (see Table 1 and Figure 1 for detailed reasons for exclusion). Therefore, even though scientific quality appraisal (scoring) was planned using Joanna Briggs Institute (JBI) evidence-based healthcare tools, this part of the analysis was not performed.

2.4. Outcomes

The primary outcome of interest was stakeholder attitudes and preferences regarding the use of Artificial Intelligence (AI) and Electronic Health Records (EHRs) in the context of rare-disease screening.

Outcomes of interest included, but were not limited to:-Perceived acceptability of these screening technologies;-Trust in AI systems;-Perceived usefulness of these technologies;-Willingness to use or adopt such technologies;-Perceived risks and ethical concerns.

3. Results

No studies met all the predefined inclusion criteria. Several articles were noted as related to (1) deep learning or algorithms; (2) detection of diagnosis, diseases, or at-risk alerts for RD; (3) the use of Electronic Health Records; and (4) consideration of attitudes or preferences of users or healthcare providers (see Table 1). Even in the shortlisted articles that underwent full-text review, most of them were excluded due to the absence of (2) detection of diagnosis, diseases, or at-risk alerts for RD. The second most frequent reason was the lack of (3) Electronic Health Records, and only 1 was excluded because it did not (4) consider attitudes or preferences of users or healthcare providers. All search results are publicly available through the OSF at https://doi.org/10.17605/OSF.IO/ZYMTW.

4. Discussion

Our findings highlight a significant gap in the literature and the limited exploration of preferences in a prolific research area. Specifically, more research is needed at the intersection of (1) deep learning or algorithms; (2) detection of diagnosis, diseases, or at-risk alerts for RD; (3) the use of Electronic Health Records; and (4) consideration of attitudes or preferences of users or healthcare providers. Given the rapid growth of technical and clinical research on the use of AI and ML with EHR data for RD screening, evidence on stakeholder perspectives is urgently needed.

Qian et al. (2021) explored the landscape and “hot topics” in the EHR literature and pointed at the phenomenal growth in publications since 2004, allowing them to analyze more than 10,000 articles [41]. They analyzed the main content topic clusters and identified key terms in the main nodes that were used in the development of this study. The authors grouped machine learning and AI under the broader heading of “technology.” Similarly, they classified genetics under “subjects.” Concepts related to attitudes and perceptions were classified as “information system researches.” Finally, patient satisfaction was categorized under “patient engagement.” This specific clustering of patient-centered and technical concepts may be a key factor in understanding the study’s findings.

References to patient preferences in the existing literature mainly focus on the technical aspect of experience or on the ethical debate surrounding AI/ML and EHR data. Consequently, the more complex topic of gNBS for RDs has not yet been explored. In terms of research subjects, “children” and “adolescents” were considered but newborns were not included. Acceptance and stakeholder preference for EHR use in the context of RDs has been examined primarily from the perspectives of data sharing and protection [42,43,44]; however, the use of AI/ML has not yet been thoroughly investigated.

If we relate our findings to the literature presented in the background, the existing evidence clearly maps the gap identified in this review. Of the systemic reviews published over the last five years, none has addressed the intersection of these domains considered in our review (see Table 2).

4.1. Identified Challenges

The lack of evidence on stakeholder perspectives and preferences regarding the use of EHR with AI or ML for identifying undiagnosed RDs may reflect the inherent complexity of both RD research and its ethical considerations. Research on RDs is challenging due to the small number of eligible participants [45] and, on a technical level, the need for large volumes of interoperable data to avoid bias in AI or ML models [46,47]. Moreover, there is an important distinction between suggesting a patient is at risk for a diagnosis and ruling out a diagnosis. This distinction may have influenced our search results, as most research do not differentiate between the terms “screening”, “testing”, or “diagnosing.” Most research focus on diagnosing symptomatic patients, whereas “flagging” or screening is a question of detecting potential risk for a disease, sometimes in the absence of classical symptoms—especially in newborns. This can be further complicated by the fact that EHR data are not specifically designed for use beyond the NBS period [48]. Similar issues exist for the terminology and usage of the key concept “AI.” The evolution and conceptualization of AI literacy is still an issue that has direct consequences for both research and society [49]. For example, AI is increasingly integrated into user-facing technologies; however, users are often told that the software they interact with relies on an “algorithm” that recognizes preferences and behavior. Such “misconceptions can limit people’s ability to effectively use, collaborate with, and act as critical consumers of AI” ([49], p1). In 2019, Wang attempted to clarify the definition of artificial intelligence beyond the classical “intelligent machine”, expanding it to encompass cognitive tasks, notably learning and problem-solving, leveraging cutting-edge technological advancements such as machine learning, natural language processing, and neural networks [50].

4.2. The Importance of “Empty Reviews”

Our results also provide an opportunity to contribute to the methodological discussion on the value of publishing null results or “empty reviews.” In the field of pharmacopsychology, it is well recognized that the non-publication of null results can distort the evidence base and introduce bias into subsequent research [51], adding to the broader “catalogue of biases” in scientific publishing [52]. At the methodological level, there is ongoing debate regarding the usefulness of “empty reviews” [53], with arguments often focusing on their prevalence within a specific field. In the fields of AI and RD, such publications are not commonly reported. On a research ethics level, while preventing publication bias is important, it is also vital to consider the need to avoid duplication of efforts [54], enhance methodological transparency [55], and thus, promote research integrity and informed decision-making [56]. Moreover, sharing null results can save other researchers time and resources, particularly in contexts where research findings are not consistently published across studies [57]. In an open data system, researchers still need to triage datasets and methods that are unlikely to yield meaningful results [58]. Highlighting the lack of research in a certain field (i.e., identifying large research gaps) can help researchers focus on areas of interest more rapidly [59]. In this context, the present study makes a methodological contribution by formalizing the use of an “empty review” in a rapidly evolving field. Such reviews can help identify neglected areas, guide research priorities, and improve transparency while reducing publication bias and duplication of effort [60,61]. This is particularly valuable in fast-moving domains such as AI in healthcare, where high publication volume can obscure important gaps in the evidence base. By clearly demonstrating a missing intersection of evidence, this review frames the absence of studies as a meaningful and actionable finding.

Moreover, the study highlights a key methodological issue: inconsistent terminology. The conflation of terms such as “screening”, “diagnosis”, and “risk detection” can obscure relevant evidence and hinder systematic retrieval. As recognized in healthcare research, such variability limits comparability and synthesis across domains [62]. Our findings suggest that part of the observed gap may stem from terminology-driven invisibility rather than a true absence of related work.

4.3. Novelty and Contribution

Existing literature has examined AI implementation, stakeholder perspectives, and EHR-based clinical applications largely in isolation, with many studies adopting a predominantly technical focus [63,64]. However, qualitative syntheses of clinical AI highlight the importance of stakeholder engagement while also identifying persistent gaps in understanding how these perspectives influence implementation outcomes [25]. Similarly, broader research on AI adoption consistently identifies trust, organizational readiness, and ethical concerns as central barriers in real-world stakeholder interaction—factors that are essential for effective implementation [64,65,66]. Yet these considerations are rarely examined in relation to specific use cases such as RD screening [67,68]. This imbalance is reflected in the broader AI literature, where evaluations tend to prioritize accuracy and efficiency over experiential, ethical, and social dimensions, and where patient perspectives remain marginal [69]. By explicitly combining AI/ML, EHR use, RD detection, and stakeholder preferences within a single framework, this review introduces a multi-dimensional gap-mapping approach that has not been systematically applied in prior work.

A second key contribution is the identification of a translational disconnect between technological development and real-world implementation. Although AI innovation is advancing rapidly, relatively few systems are successfully integrated into routine clinical practice [25,68,70]. In the RD context, AI has been framed as transformative across the patient journey; however, this vision presumes alignment with stakeholder expectations that remains underexplored [71]. This review provides structured evidence that such alignment is currently lacking, extending the concept of the research-to-practice gap to include stakeholder-informed AI deployment.

4.4. Limitations

The findings of this systematic review should be interpreted in light of several methodological considerations that may partly explain the absence of eligible studies. While our results suggest a lack of evidence at the intersection of AI/ML, EHR use, RD detection, and stakeholder preferences, this may also reflect potential limitations in evidence retrieval. Terminology heterogeneity could have played a significant role, as concepts in this field—such as “screening”, “early detection”, “phenotyping”, “flagging”, and “diagnosis”—are often used interchangeably across disciplines. This variability may have reduced the sensitivity of the search strategy across databases. Similarly, variability in how “artificial intelligence” is described (e.g., algorithms, machine learning, decision support systems) may have contributed to incomplete retrieval, despite efforts to use broad search terms. Moreover, the review was restricted to English-language publications, which may have introduced language bias [72,73] particularly given the global nature of rare-disease research and AI development. Although major databases were included, relevant studies indexed in specialized technical repositories (e.g., computer science databases) or other interdisciplinary sources may not have been captured. Finally, the exclusion of grey literature—often omitted in SLR in the medical field—may have limited the scope of identified evidence, particularly in the field of AI, where innovations are often first reported outside traditional academic publishing channels.

4.5. Future Research Directions

This review highlights an urgent need for empirical, stakeholder-centered research at the intersection of AI/ML, EHR-based screening, and rare diseases (RDs). Future work should move beyond technical performance to examine the perspectives of patients, caregivers, and healthcare providers, particularly in pre-diagnostic contexts marked by high uncertainty—areas still underrepresented in current research and governance frameworks [72,74,75]. Priority areas include: (1) acceptance thresholds for risk-flagging versus formal diagnosis, (2) preferences regarding secondary use of EHR data, consent models, and data governance, and (3) the perceived psychosocial impact of being identified as “at risk” for an RD. Given that willingness to share health data varies by context and is shaped by trust, transparency, and perceived benefit [23], further research should also examine attitudes toward data reuse, linkage, and cross-institutional sharing, alongside concerns about privacy, representativeness, and data quality [76]. Mixed-methods and participatory study designs will be essential, with explicit comparisons across stakeholder groups.

In parallel, future research should explicitly integrate principles of explainable AI (XAI) into preference studies. The “black-box” nature of many AI systems remains a major barrier to trust and adoption, and explainability has been identified as critical for improving interpretability, accountability, and clinical usability [77,78]. Empirical studies should assess how transparency and communication of uncertainty influence trust, particularly in EHR-based systems that rely on complex and imperfect data. Testing different explanation formats (e.g., feature attribution methods, narratives, and probabilistic outputs) and varying levels of detail will be crucial to determine what is meaningful and acceptable to diverse users [74,79]. Embedding XAI within user-centered design approaches is essential to ensure that EHR-based AI tools are not only accurate but also interpretable, trustworthy, and aligned with stakeholder expectations.

5. Conclusions

In conclusion, this study highlights a significant gap in the literature regarding stakeholder preference studies in the context of RD screening and, consecutively, the use of EHR data for RD detection). This gap stands in contrast to the extensive body of research on technical and clinical aspects of AI and ML applications. The limited focus on stakeholder preferences in EHR-based AI use underscores the methodological complexity and ethical challenges inherent in this area. This gap is further compounded by the small number of eligible participants in RD research and the inconsistencies in terminology, particularly the overlap between screening and diagnosis. Additionally, the publication of null results, as demonstrated by our findings, is crucial for enhancing methodological transparency, reducing publication bias, and promoting research integrity. Addressing these gaps and challenges is essential to advance the field and support informed decision-making in the development and implementation of emerging healthcare technologies.

Acknowledgements

AcknowledgmentsWe thank the Uppsala Librarian Team for their support in developing and refining the search strategy.

Competing interests

Conflicts of InterestAuthor Josh Coulter was employed by Pfizer, and authors Alain Anyouzoa and Kaisa Elomaa were employed by Takeda Oy. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflicts of interest.

References

1. PREFER
2. Tingley K., Coyle D., Graham I.D., Chakraborty P., Wilson K., Potter B.K.. Stakeholder perspectives on clinical research related to therapies for rare diseases: Therapeutic misconception and the value of research. Orphanet J. Rare Dis.. 2021 doi:10.1186/s13023-020-01624-0 PubMed
3. Yoon S., Lee M., Jung H.-I., Khan M.M., Kim S.-Y., Kim H. et al. Prioritization of research engaged with rare disease stakeholders: A systematic review and thematic analysis. Orphanet J. Rare Dis.. 2023 doi:10.1186/s13023-023-02892-2 PubMed
4. Gittus M., Chong J., Sutton A., Ong A.C., Fotheringham J.. Barriers and facilitators to the implementation of guidelines in rare diseases: A systematic review. Orphanet J. Rare Dis.. 2023 doi:10.1186/s13023-023-02667-9 PubMed
5. Fujita H.. AI-based computer-aided diagnosis (AI-CAD): The latest review to read first. Radiol. Phys. Technol.. 2020 doi:10.1007/s12194-019-00552-4 PubMed
6. List of Rare Diseases
7. Sun K., Roy A., Tobin J.M.. Artificial intelligence and machine learning: Definition of terms and current concepts in critical care research. J. Crit. Care. 2024 doi:10.1016/j.jcrc.2024.154792 PubMed
8. Lin S., Nateqi J., Weingartner-Ortner R., Gruarin S., Marling H., Pilgram V. et al. An artificial intelligence-based approach for identifying rare disease patients using retrospective electronic health records applied for Pompe disease. Front. Neurol.. 2023 doi:10.3389/fneur.2023.1108222 PubMed
9. Wojtara M., Rana E., Rahman T., Khanna P., Singh H.. Artificial intelligence in rare disease diagnosis and treatment. Clin. Transl. Sci.. 2023 doi:10.1111/cts.13619 PubMed
10. Holm I.A., Agrawal P.B., Ceyhan-Birsoy O., Christensen K.D., Fayer S., Frankel L.A. et al. The BabySeq project: Implementing genomic sequencing in newborns. BMC Pediatr.. 2018 doi:10.1186/s12887-018-1200-1 PubMed
11. Ferlini A., Gross E.S., Garnier N.. Rare diseases’ genetic newborn screening as the gateway to future genomic medicine: The Screen4Care EU-IMI project. Orphanet J. Rare Dis.. 2023 doi:10.1186/s13023-023-02916-x PubMed
12. Lehne M., Sass J., Essenwanger A., Schepers J., Thun S.. Why digital medicine depends on interoperability. npj Digit. Med.. 2019 doi:10.1038/s41746-019-0158-1 PubMed
13. Lochmüller H., Badowska D.M., Thompson R., Knoers N.V., Aartsma-Rus A., Gut I. et al. RD-Connect, NeurOmics and EURenOmics: Collaborative European initiative for rare diseases. Eur. J. Hum. Genet.. 2018 doi:10.1038/s41431-018-0115-5 PubMed
14. DeCamp M., Lindvall C.. Latent bias and the implementation of artificial intelligence in medicine. J. Am. Med. Inform. Assoc.. 2020 doi:10.1093/jamia/ocaa094 PubMed
15. Kim E., Rubinstein S.M., Nead K.T., Wojcieszynski A.P., Gabriel P.E., Warner J.L.. The evolving use of electronic health records (EHR) for research. Semin. Radiat. Oncol.. 2019 doi:10.1016/j.semradonc.2019.05.010 PubMed
16. Ebnehoseini Z., Tara M., Tabesh H., Dindar F.H., Hasibian S.. Understanding key factors affecting on hospital electronic health record (EHR) adoption. J. Fam. Med. Prim. Care. 2020 doi:10.4103/jfmpc.jfmpc_109_20 PubMed
17. Buendia O., Shankar S., Mahon H., Toal C., Menzies L., Ravichandran P. et al. Is it possible to implement a rare disease case-finding tool in primary care? A UK-based pilot study. Orphanet J. Rare Dis.. 2022 doi:10.1186/s13023-022-02216-w PubMed
18. Garcelon N., Burgun A., Salomon R., Neuraz A.. Electronic health records for the diagnosis of rare diseases. Kidney Int.. 2020 doi:10.1016/j.kint.2019.11.037 PubMed
19. Kim S., Zhou Y., Guo Y., Xiao C., Zheng K.. Applying natural language processing and large language models to clinical notes for phenotyping and diagnosing rare diseases: A systematic review. J. Am. Med. Inform. Assoc.. 2026 doi:10.1093/jamia/ocag045 PubMed
20. Nishat S.M.H., Tanweer A.S., Alshamsi B., Shaheen M.H., Tanveer A.S., Nishat A. et al. Artificial intelligence: A new frontier in rare disease early diagnosis. Cureus. 2025 doi:10.7759/cureus.79487 PubMed
21. Yang S., Varghese P., Stephenson E., Tu K., Gronsbell J.. Machine learning approaches for electronic health records phenotyping: A methodical review. J. Am. Med. Inform. Assoc.. 2023 doi:10.1093/jamia/ocac216 PubMed
22. Dlima S.D., Shevade S., Menezes S.R., Ganju A.. Digital Phenotyping in Health Using Machine Learning Approaches: Scoping Review. JMIR Bioinform. Biotech.. 2022 doi:10.2196/39618 PubMed
23. Cascini F., Pantovic A., Al-Ajlouni Y.A., Puleo V., De Maio L., Ricciardi W.. Health data sharing attitudes towards primary and secondary use of data: A systematic review. eClinicalMedicine. 2024 doi:10.1016/j.eclinm.2024.102551 PubMed
24. Alam P., Bolio A., Lin L., Larson H.J.. Stakeholders’ perceptions of personal health data sharing: A scoping review. PLoS Digit. Health. 2024 doi:10.1371/journal.pdig.0000652 PubMed
25. Hogg H.D.J., Al-Zubaidy M., Talks J., Denniston A.K., Kelly C.J., Malawana J. et al. Stakeholder Perspectives of Clinical Artificial Intelligence Implementation: Systematic Review of Qualitative Evidence. J. Med. Internet Res.. 2023 doi:10.2196/39742 PubMed
26. Ling Kuo R.Y., Freethy A., Smith J., Hill R., C J., Jerome D. et al. Stakeholder perspectives towards diagnostic artificial intelligence: A co-produced qualitative evidence synthesis. eClinicalMedicine. 2024 doi:10.1016/j.eclinm.2024.102555 PubMed
27. Tun H.M., Rahman H.A., Naing L., Malik O.A.. Trust in Artificial Intelligence–Based Clinical Decision Support Systems Among Health Care Workers: Systematic Review. J. Med. Internet Res.. 2025 doi:10.2196/69678 PubMed
28. Cochrane
29. Poulsen W., Christensen K., Dalgård C.. Dietary patterns and survival to 100+ years: An empty systematic review of cohort and case-control studies. Arch. Public Health. 2022 doi:10.1186/s13690-022-00914-2 PubMed
30. Albrink K., Schröder D., Joos C., Müller F., Noack E.M.. Usability of an app for medical history taking in general practice from the patients’ perspective: Cross-sectional study. JMIR Hum. Factors. 2024 doi:10.2196/47755 PubMed
31. Funes Hernandez M., Babakhanian M., Chen T.P., Sarraju A., Seninger C., Ravi V. et al. Design and implementation of an electronic health record-integrated hypertension management application. J. Am. Heart Assoc.. 2024 doi:10.1161/JAHA.123.030884 PubMed
32. Jauk S., Kramer D., Avian A., Berghold A., Leodolter W., Schulz S.. Technology acceptance of a machine learning algorithm predicting delirium in a clinical setting: A mixed-methods study. J. Med. Syst.. 2021 doi:10.1007/s10916-021-01727-6 PubMed
33. Jones K.H., Daniels H., Squires E., Ford D.V.. Public views on models for accessing genomic and health data for research: Mixed methods study. J. Med. Internet Res.. 2019 doi:10.2196/14384 PubMed
34. Lærum H., Bremer S., Bergan S., Grünfeld T.. A taste of individualized medicine: Physicians’ reactions to automated genetic interpretations. J. Am. Med. Inform. Assoc.. 2014 doi:10.1136/amiajnl-2012-001587 PubMed
35. Mo H., Thompson W.K., Rasmussen L.V., Pacheco J.A., Jiang G., Kiefer R. et al. Desiderata for computable representations of electronic health records-driven phenotype algorithms. J. Am. Med. Inform. Assoc.. 2015 doi:10.1093/jamia/ocv112 PubMed
36. Pathak J., Kho A.N., Denny J.C.. Electronic health records-driven phenotyping: Challenges, recent advances, and perspectives. J. Am. Med. Inform. Assoc.. 2013 doi:10.1136/amiajnl-2013-002428 PubMed
37. Richesson R.L., Hammond W.E., Nahm M., Wixted D., Simon G.E., Robinson J.G. et al. Electronic health records based phenotyping in next-generation clinical trials: A perspective from the NIH Health Care Systems Collaboratory. J. Am. Med. Inform. Assoc.. 2013 doi:10.1136/amiajnl-2013-001926 PubMed
38. Sewitch M.J., Hilsden R., Joseph L., Rabeneck L., Paszat L., Bitton A. et al. Qualitative study of physician perspectives on classifying screening and nonscreening colonoscopy using administrative health data: Adding practice does not make perfect. Can. J. Gastroenterol. Hepatol.. 2012 doi:10.1155/2012/176714 PubMed
39. Smith S., Trevithick R.W., Smith J., Pung L., Taylor K., Ha N. et al. “Currently flying blind” Stakeholders’ perceptions of implementing statewide population-based cancer staging at diagnosis into the Western Australian Cancer Registry: A rapid qualitative process evaluation of the WA Cancer Staging Project. BMC Health Serv. Res.. 2023 doi:10.1186/s12913-023-09662-7 PubMed
40. Zhang Z., Genc Y., Wang D., Ahsen M.E., Fan X.. Effect of AI explanations on human perceptions of patient-facing AI-powered healthcare systems. J. Med. Syst.. 2021 doi:10.1007/s10916-021-01743-6 PubMed
41. Qian Y., Ni Z., Gui W., Liu Y.. Exploring the Landscape, Hot Topics, and Trends of Electronic Health Records Literature with Topics Detection and Evolution Analysis. Int. J. Comput. Intell. Syst.. 2021 doi:10.2991/ijcis.d.210203.006
42. Courbier S., Dimond R., Bros-Facer V.. Share and protect our health data: An evidence based approach to rare disease patients’ perspectives on data sharing and data protection-quantitative survey and recommendations. Orphanet J. Rare Dis.. 2019 doi:10.1186/s13023-019-1123-4 PubMed
43. Raspa M., Paquin R.S., Brown D.S., Andrews S., Edwards A., Moultrie R. et al. Preferences for accessing electronic health records for research purposes: Views of parents who have a child with a known or suspected genetic condition. Value Health. 2020 doi:10.1016/j.jval.2020.06.016 PubMed
44. Baines R., Stevens S., Garba-Sani Z., Chatterjee A., Austin D., Leigh S.. Attitudes Toward Personal Health Data Sharing Among People Living With Sickle Cell Disorder, Exemplar for Study of Rare Disease Populations. J. Patient-Cent. Res. Rev.. 2023 doi:10.17294/2330-0698.2006 PubMed
45. Whicher D., Philbin S., Aronson N.. An overview of the impact of rare disease characteristics on research methodology. Orphanet J. Rare Dis.. 2018 doi:10.1186/s13023-017-0755-5 PubMed
46. Li X., Yu L., Jin Y., Fu C.-W., Xing L., Heng P.-A.. Difficulty-aware meta-learning for rare disease diagnosis. Proceedings of the Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, 4–8 October 2020. 2020 doi:10.1007/978-3-030-59710-8_35
47. Li X., Yu L., Jin Y., Fu C.-W., Xing L., Heng P.-A.. Rare disease classification via difficulty-aware meta learning. Meta Learning with Medical Imaging and Health Informatics Applications. 2023
48. Gilbert R., Logan S., Moyer V.A., Elliott E.J.. Evidence-Based Case Review: Assessing diagnostic and screening tests: Part 1. Concepts. West. J. Med.. 2001 doi:10.1136/ewjm.174.6.405 PubMed
49. Long D., Magerko B.. What is AI literacy? Competencies and design considerations. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems. 2020 doi:10.1145/3313831.3376727
50. Wang P.. On defining artificial intelligence. J. Artif. Gen. Intell.. 2019 doi:10.2478/jagi-2019-0002
51. Campbell K.L., Moore J.B., Bartholomew J.B.. The importance of publishing null results: Editorial guidelines to contribute to the reduction of publication bias in translational exercise research. Transl. J. Am. Coll. Sports Med.. 2020 doi:10.1249/tjx.0000000000000141
52. DeVito N.J., Goldacre B.. Catalogue of bias: Publication bias. BMJ Evid.-Based Med.. 2019 doi:10.1136/bmjebm-2018-111107 PubMed
53. Gray R.. Empty systematic reviews: Identifying gaps in knowledge or a waste of time and effort?. Nurse Author Ed.. 2021 doi:10.1111/nae2.23
54. Baxter M.G., Burwell R.D.. Promoting transparency and reproducibility in Behavioral Neuroscience: Publishing replications, registered reports, and null results. Behav. Neurosci.. 2017 doi:10.1037/bne0000207 PubMed
55. Marsden E.. Methodological transparency and its consequences for the quality and scope of research. The Routledge Handbook of Research Methods in Applied Linguistics. 2019
56. Bespalov A., Steckler T., Skolnick P.. Be positive about negatives–recommendations for the publication of negative (or null) results. Eur. Neuropsychopharmacol.. 2019 doi:10.1016/j.euroneuro.2019.10.007 PubMed
57. Scherer R.W., Ugarte-Gil C., Schmucker C., Meerpohl J.J.. Authors report lack of time as main reason for unpublished research presented at biomedical conferences: A systematic review. J. Clin. Epidemiol.. 2015 doi:10.1016/j.jclinepi.2015.01.027 PubMed
58. Borycz J., Olendorf R., Specht A., Grant B., Crowston K., Tenopir C. et al. Perceived benefits of open data are improving but scientists still lack resources, skills, and rewards. Hum. Soc. Sci. Comm.. 2023 doi:10.1057/s41599-023-01831-7
59. Baako I., Alhassan H., Gidisu P.. Understanding and spotting research gaps through a systematic literature review. Int. J. Res. Innov. Soc. Sci.. 2022
60. Moreal C., Magro G., D’Angela D., Moreale R., Dussi G., Dentice S. et al. Publication Dynamics Where Evidence Is Missing: Mapping Empty Reviews in Nursing. Worldviews Evid.-Based Nurs.. 2026 doi:10.1111/wvn.70125 PubMed
61. Yaffe J., Montgomery P., Hopewell S., Shepard L.D.. Empty Reviews: A Description and Consideration of Cochrane Systematic Reviews with No Included Studies. PLoS ONE. 2012 doi:10.1371/journal.pone.0036626 PubMed
62. Breit H., Panos N.G., Fisher K., Blackwood A., Yeager C., Schachter M. et al. Delirium screening in the neurointensive care unit: Comparison of the confusion assessment method for the intensive care unit and intensive care delirium screening checklist. J. Neurol. Sci.. 2025 doi:10.1016/j.jns.2025.123619 PubMed
63. Kaye J., Shah N., Kogetsu A., Coy S., Katirai A., Kuroda M. et al. Moving beyond Technical Issues to Stakeholder Involvement: Key Areas for Consideration in the Development of Human-Centred and Trusted AI in Healthcare. Asian Bioeth. Rev.. 2024 doi:10.1007/s41649-024-00300-w PubMed
64. Asan O., Choudhury A.. Research Trends in Artificial Intelligence Applications in Human Factors Health Care: Mapping Review. JMIR Hum. Factors. 2021 doi:10.2196/28236 PubMed
65. Norori N., Hu Q., Aellen F.M., Faraci F.D., Tzovara A.. Addressing bias in big data and AI for health care: A call for open science. Patterns. 2021 doi:10.1016/j.patter.2021.100347 PubMed
66. Hicks S.A., Strümke I., Thambawita V., Hammou M., Riegler M.A., Halvorsen P. et al. On evaluation metrics for medical applications of artificial intelligence. Sci. Rep.. 2022 doi:10.1038/s41598-022-09954-8 PubMed
67. Ahmed M.I., Spooner B., Isherwood J., Lane M., Orrock E., Dennison A.. A Systematic Review of the Barriers to the Implementation of Artificial Intelligence in Healthcare. Cureus. 2023 doi:10.7759/cureus.46454 PubMed
68. Nair M., Svedberg P., Larsson I., Nygren J.M.. A comprehensive overview of barriers and strategies for AI implementation in healthcare: Mixed-method design. PLoS ONE. 2024 doi:10.1371/journal.pone.0305949 PubMed
69. Wang Z., Yan R., Francis S., Diaz C., Flickinger T., Lin Y. et al. Stakeholder-centric participation in large language models enhanced health systems. npj Health Syst.. 2025 doi:10.1038/s44401-025-00024-5 PubMed
70. McCradden M.D., Joshi S., Anderson J.A., London A.J.. A normative framework for artificial intelligence as a sociotechnical system in healthcare. Patterns. 2023 doi:10.1016/j.patter.2023.100864 PubMed
71. Groza T., Baynam G., Jamuar S.S.. Reimagining care of people living with rare diseases with artificial intelligence. PLoS Med.. 2026 doi:10.1371/journal.pmed.1004966 PubMed
72. Ndong A.. Language and geographical bias limits global health research. BMJ. 2026 doi:10.1136/bmj.s436 PubMed
73. Pieper D., Puljak L.. Language restrictions in systematic reviews should not be imposed in the search strategy but in the eligibility criteria if necessary. J. Clin. Epidemiol.. 2021 doi:10.1016/j.jclinepi.2020.12.027 PubMed
74. Hou J., Zhang Z., Cheng X., Wang W.. Patient Concerns Regarding Artificial Intelligence Applications in Health Care: Systematic Review and Meta-Synthesis Based on Social Ecological Theory. J. Med. Internet Res.. 2026 doi:10.2196/85663 PubMed
75. Nong P., Ji M.. Expectations of healthcare AI and the role of trust: Understanding patient views on how AI will impact cost, access, and patient-provider relationships. J. Am. Med. Inform. Assoc.. 2025 doi:10.1093/jamia/ocaf031 PubMed
76. Pant D., Nytrø Ø., Leventhal B.L., Clausen C., Koochakpour K., Stien L. et al. Secondary use of health records for prediction, detection, and treatment planning in the clinical decision support system: A systematic review. BMC Med. Inform. Decis. Mak.. 2025 doi:10.1186/s12911-025-03021-8 PubMed
77. Agbo Eje O., Azim S.M., Dehzangi I.. Explainable AI Applications in Healthcare: A Systematic Review. Algorithms. 2026 doi:10.3390/a19060488
78. Salimparsa M., Sedig K., Lizotte D.J., Abdullah S.S., Chalabianloo N., Muanda F.T.. Explainable AI for Clinical Decision Support Systems: Literature Review, Key Gaps, and Research Synthesis. Informatics. 2025 doi:10.3390/informatics12040119
79. Nong P., Platt J.. Patients’ Trust in Health Systems to Use Artificial Intelligence. JAMA Netw. Open. 2025 doi:10.1001/jamanetworkopen.2024.60628 PubMed